Scriba
Local speaker-diarized meeting transcription (whisperX + pyannote), then rename Speaker 1/2 to real names.
Działa po konfiguracji
Co robi ten skill
Transcribes an audio/video meeting into an accurate, speaker-labeled Markdown transcript (Speaker 1/2/3) with embedded per-speaker voice clips, then renames speakers to real names on request. Triggers when the user drops an audio/video file and asks for a transcript, diarization, or 'who said what' (English or Russian).
Raport z testu
Ran its own real formatter end-to-end on a synthetic diarized transcript: correct 62%/38% talk-time math, clean speaker-ID section with sample quotes, then a working rename pass (Speaker 1/2 -> real names) -- genuinely better than a naive baseline render. Full ASR+diarization couldn't be exercised: it needs `uv` (not installed here) plus a free HuggingFace account and a 3-click gated-model consent flow before a ~3GB download even starts, exactly as SKILL.md itself discloses.
Testowano: 2026-07-15 · Claude Code 2.x (agent harness)
Instalacja
git clone https://github.com/AlexanderAbramovPav/scriba mkdir -p ~/.claude/skills cp -r scriba ~/.claude/skills/scriba
Komendy i przykładowe prompty
/scribaLocal speaker-diarized meeting transcription (whisperX + pyannote), then rename Speaker 1/2 to real names.
Skille uruchamiają się na zwykłe polecenia — bez komend do zapamiętania. Po instalacji aktywują go prompty takie jak te (po angielsku):
Transcribe this meeting recording and label who said what, pleaseDiarize this audio file and then rename the speakers to their real namesI have a Zoom recording, give me a speaker-labeled transcript please